Three-dimensional convolutional neural network (3D-CNN) for satellite behavior discovery

Abstract

Space superiority includes space protection and space situational awareness (SSA), which require rapid and accurate space object behavioral motion and operational intent discovery. The presence of clutter, in addition to real-time and hidden information constraints, greatly complicates the space awareness decision-making to control both ground-based and space-based surveillance assets. Space is considered as an important concern in modern frontiers because intelligence information from the space has become extremely vital for strategic decisions, which calls for persistent Space Domain Awareness (SDA). The presence of disagreeable actors in addition to real-time and hidden information constraints greatly complicates the decision-making process in satellite behavior detection as well as operational intent discovery. This paper designs and implements 3D-Convoltional Neural Networks (CNNs) for rapid discovery of evasive satellite behaviors from ground-based sensors, which measure the ranges, azimuth angles, and elevation angles in the Adaptive Markov Inference Game Optimization (AMIGO) tool. The novel 3D CNN extends the generic 2d CNN towards analysis from many perspectives. To generate the 3D CNN model, the training and validation data are simulated based on our game theoretic reasoning engine for elusive space behaviors detection, interactive adversary awareness, and intelligent probing. The performance of the 3D CNN is compared with the 2D CNN models from previous work which is shown for a 10% increase in accuracy.